Projects
Things I build for myself, usually because I went looking for them and they didn't exist.
sarvis
Behavioral intelligence, quietly running in the background.
Why I built it
I noticed I kept repeating the same mistakes. Not big ones, small ones. Patterns I couldn't see because I was too close to them. I wanted something that watched over time and surfaced those patterns without me having to go looking. Not a journal. Not a mood tracker. Something quieter.
What it is
sarvis captures raw inputs throughout the day (notes, thoughts, tasks), structures them via LLM, and turns them into behavioral signals over time. Built in Swift. The core pipeline (capture, LLM structuring, pattern extraction) is working. The hard part isn't the engineering, it's figuring out what a useful "pattern" actually is.
cortex
A PM brain for any git repo.
Why I built it
Every time I started a new session with an AI coding assistant, I'd spend the first ten minutes re-explaining the project. What it does, how it's structured, what's already been tried. That's wasted time every single session. I wanted the assistant to already know. So I made a way for it to know.
What it is
cortex is a Claude Code skill I built and open-sourced. It installs a post-commit hook that automatically scans your repo and updates a private context file after every push. Any AI assistant you bring into the project picks it up immediately. No re-explaining.
→ GitHubWikipedia Continual-Learning RAG
A RAG system that grounds itself on Wikipedia and fine-tunes itself when it can't find an answer. It keeps learning instead of going stale.
PythonML Model to Production on GKE
An end-to-end deployment pipeline on GCP: train, containerize with Docker, push to Artifact Registry, deploy on GKE, serve through a frontend. Written up step by step on Medium.
Python · GCPSpotify Recommendation System
Pulls listening data from the Spotify API, lands it in Cloud Storage, orchestrates with Airflow into MongoDB, and serves recommendations on top.
Python · Airflow